[PD-dev] First complete keyboard navigation prototype

Tue Aug 20 05:07:48 CEST 2019

Thanks a lot for your input Christof!

If i may let me ask a couple of question:

1) I've been wandering through the sources and i'm not sure: where should i put the "#def HAVE_KEYBOARDNAV"? Also this value will be set manually in the code or is there anyway to set it via some config file/some other way?

_______________________________________

2) Also i've put a big portion of the code in a g_kbdnav.c file a long with it's forward declarations in g_kbdnav.h.

Why it was in g_editor? Because it uses canvas_connect_with_undo, for example, which is static.

What should i do? Change those methods so they're not static anymore?

Or maybe put a "helper function" inside g_editor.c?

in g_editor.c:

void kbdnav_connect_with_undo(t_canvas *x, t_float index1, t_float outno, t_float index2, t_float inno){
    connect_with_undo(x, index1,outno, index2, inno);
}

in g_kbdnav.c:

[...]
kbdnav_connect_with_undo(...)
[...]

Or should i do something else?
_______

By the way i wanted to hear more from you guys about the experience when trying the kbdnav prototype! :)
Almost no one commented.

Best,
Henri.

________________________________
From: Christof Ressi <christof.ressi at gmx.at>

Hi,

> But it appears that actually the size of the struct is deduced while compiling. So if you write an external it is going to think the canvas struct is the same size as it was in the pd header files that were used during compilation.

exactly.

> When using the pointer to implementation idiom, wouldn't the pointer itself change the size of the struct?

yes, it will

> You've said something about that not being a problem when you add it as the last member of the struct but i don't understand why and how that would work.

usually, externals shouldn't care about the *size* the t_editor struct (at least I can't think of any use case), so you can get away with adding new fields at the end (although it's certainly not recommended). Note that those headers aren't really public anyway!

However, appending fields conditionally can lead to problems:

struct Foo {
    int a;
#ifdef FOO_EX
    int c;
#endif
};

Now let's say we need to add another member:

struct Foo {
    int a;
#ifdef FOO_EX
    int c;
#endif
    in b;
};

If the host compiles with FOO_EX defined and the client doesn't, the latter will assume a wrong offset for 'b'.

The solution is to add a field for private data *once*. The advantage is that a) we can hide the private data and b) we can extend it without worrying about compatibility:

struct Foo {
   int a;
   PrivateFoo *p;
};

We can still add public members if needed:

struct Foo {
    int a;
    void *private;
    int b;
};

'private' points to a private data structure that is not be visible to clients. There you can conditionally enable members without problems:

struct PrivateFoo {
#ifdef USE_BAR
    struct MyFeature feature;
#endif
};

MyFeature could be in a seperate source file together with your methods and it only gets compiled when needed.

Again, have a look at the "t_canvas_private" struct and the "gl_privatedata" member of "_glist" (aka "t_canvas") and do the same for "_editor", e.g.:

in g_canvas.h:

typedef struct _editor {
    ...
    void *e_privatedata;
} t_editor;

in g_editor.c:

#ifdef HAVE_KEYBOARDNAV
#include "g_keyboardnav.h"
#endif

typedef struct _editor_private {
#ifdef HAVE_KEYBOARDNAV
    t_keyboardnav keyboardnav;
#endif
} t_editor_private;

the "t_keyboardnav" struct is defined in "g_keyboardnav.h" and its methods implemented in "g_keyboardnav.c". Both only get compiled when needed.

Hope this makes sense.

Christof

Gesendet: Dienstag, 16. Juli 2019 um 16:51 Uhr
Von: "Henri Augusto Bisognini" <msndohenri at hotmail.com>
An: "pd-dev at lists.iem.at" <pd-dev at lists.iem.at>
Betreff: Re: [PD-dev] First complete keyboard navigation prototype
Okay, just help me check if i got it right.

At first i was thinking that when externals, for any reason, used the size of the canvas struct (or any other) it would do so in real time. Like calling sizeof() and stuff.

But it appears that actually the size of the struct is deduced while compiling. So if you write an external it is going to think the canvas struct is the same size as it was in the pd header files that were used during compilation.

Is that it?

I don't think i quite get something. When using the pointer to implementation idiom, wouldn't the pointer itself change the size of the struct? You've said something about that not being a problem when you add it as the last member of the struct but i don't understand why and how that would work.

(I don't have a formal background on programming so forgive me if thats obvious or something.)

________________________________
De: Christof Ressi <christof.ressi at gmx.at>
Enviado: sábado, 15 de junho de 2019 17:11
Para: Henri Augusto Bisognini
Cc: pd-dev at lists.iem.at
Assunto: Aw: Re: [PD-dev] First complete keyboard navigation prototype

Hi, as IOhannes said, "g_canvas.h" is semi-public in a sense that some externals use it (e.g. iemguts). So unless it is absolutely necessary, we should avoid breaking binary compatibility.

If the e_kbdnav member is only conditionally enabled with an #ifdef, existing externals (or those not compiled with key-nav-support) will see a different size of t_editor. This is not much of a problem as long as e_kbdnav is the last member of t_editor, but as soon as we add another member, we might run into problems, since this last field will be at a different offset.

I think the solution is simple: just add a "void *e_private" member which points to some private data where we can put all stuff which we don't want to expose the header. (This is called the "PIMPL idiom"). e_kbdnav would be the first member of such private data.

IOhannes actually did this with the "gl_privatedata" member in t_canvas to hide the undo queue implemention.  The "t_canvas_private" struct currently only has a "t_undo" member but it's possible to add/remove/rearrange members at will without having to think about binary compatibility issues because it's not in a header file.

Christof
Gesendet: Samstag, 15. Juni 2019 um 19:58 Uhr
Von: "Henri Augusto Bisognini" <msndohenri at hotmail.com>
An: "pd-dev at lists.iem.at" <pd-dev at lists.iem.at>
Betreff: Re: [PD-dev] First complete keyboard navigation prototype
Please excuse my ignorance on that matter but could you give me a brief explanation of the problem at hand?

________________________________
De: Pd-dev <pd-dev-bounces at lists.iem.at> em nome de IOhannes m zmölnig <zmoelnig at iem.at>
Enviado: sexta-feira, 14 de junho de 2019 04:37
Para: pd-dev at lists.iem.at
Assunto: Re: [PD-dev] First complete keyboard navigation prototype

On 6/13/19 7:34 PM, Henri Augusto Bisognini wrote:
> Also, in g_canvas.h, inside the "struct _editor" there is a "struct _kbdnav" member. This is the struct that contains the data used in the keyboard navigation.
>

i haven't looked at the actual code, but what you describe here, is that
you are actually changing the size of a quasi-public struct and thus the
memory layout as presented to externals.

which means that a version of Pd that has the keyboard-navigation
enabled is (partly) *binary-incompatible* with a version of Pd that does
not have the keyboard-navigation enabled.

bummer :-(

gfmadr
IOhannes

_______________________________________________ Pd-dev mailing list Pd-dev at lists.iem.at https://lists.puredata.info/listinfo/pd-dev
_______________________________________________ Pd-dev mailing list Pd-dev at lists.iem.at https://lists.puredata.info/listinfo/pd-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puredata.info/pipermail/pd-dev/attachments/20190820/b00947c1/attachment-0001.html>